Beam-loss detection for the high-rate superconducting upgradeto the SLAC Linac Coherent Light Source

Alan S. Fisher ,* Christine I. Clarke, Bryce T. Jacobson, Ruslan Kadyrov , Evan Rodriguez,Mario Santana Leitner , Leonid Sapozhnikov , and James J. WelchSLAC National Accelerator Laboratory, Menlo Park, California 94025, USA

(Received 18 March 2020; accepted 16 July 2020; published 14 August 2020)

The Linac Coherent Light Source (LCLS) x-ray free-electron laser is driven by the third kilometer of the3-km SLAC linac, which accelerates electrons in normal-conducting copper cavities pulsed at 120 Hz. Thefirst kilometer is being replaced by LCLS-II, a superconducting (SC) electron linac driven by continuous rfat 1.3 GHz and with a normal-conducting photocathode gun using continuous rf at a subharmonic,186 MHz. Its 4-GeV, 120-kW beam has a 1-MHz maximum rate, with an upgrade to 8 GeV in planning.The beam from either linac can be switched pulse by pulse to either of two new undulators, to generate hardand soft x rays. Control of beam loss is critical for machine and personnel safety. Previous SLAC protectionsystems have depended on ionization chambers, including both local devices at expected loss sites and longgas-dielectric coaxial cables providing distributed coverage. These devices are unsuited to the SC-linacbeam, because their ion collection time, over 1 ms, may allow the space charge of accumulated ions to nullthe electric field inside the detector, blinding it to an increase in loss. Instead, both the local and thedistributed detectors have been replaced with faster devices. The full 4 km will be spanned by multipleradiation-hard optical fibers in lengths of up to 200 m, each coupled to a photomultiplier tube, to captureCherenkov light from loss showers. These are supplemented by single-crystal diamond detectors atexpected loss sites. Signals are integrated with a 500-ms time constant; the beam is stopped within 200 μs ifa threshold is exceeded. We report on our extensive tests of the detectors and the new signal processing.

DOI: 10.1103/PhysRevAccelBeams.23.082802

I. INTRODUCTION

The Linac Coherent Light Source (LCLS) x-ray free-electron laser (FEL) at SLAC National AcceleratorLaboratory began operation in 2009 [1] using the thirdkilometer of SLAC’s 3-km linear accelerator (Fig. 1). Itretained the 2856-MHz normal-conducting (NC) copperlinac used in all projects since 1966 but added a 1.6-cellcopper radio-frequency (rf) photocathode gun at 2856 MHzand two bunch-compression chicanes. Both the gun and thelinac are pulsed at 120 Hz.For the LCLS-II project [2], SLAC has removed the first

kilometer of the linac, completely emptying this part of thetunnel and the klystron gallery above it for the first timesince construction. Cryomodules for a new superconduct-ing (SC) linac are being installed to replace the copperstructures. Commissioning is planned for 2021. In addition,

two variable-gap undulators, for hard and soft x rays(Fig. 1), have replaced the fixed-gap undulator used forthe past decade. Kickers will direct beams from either linacinto either undulator, or into a dump upstream, providingusers with full flexibility in the x-ray energy and rate.FACET-II, a user facility mainly for advanced acceler-

ation studies [3], will occupy the middle kilometer with acopper linac and rf photocathode gun, both pulsed at 30 Hz.The beam from the LCLS SC linac will bypass bothFACET-II and the LCLS NC linac in a transport linesuspended from the tunnel ceiling, only 125 cm from thetwo linacs below.Table I compares some parameters of the superconducting

and copper linacs. The SC linac, driven with continuous-wave (cw) 1.3-GHz rf, accelerates electrons to 4 GeV. Itsnormal-conducting photocathodegun is drivenby cw rf at theseventh subharmonic (185.7 MHz) of the linac frequency.The pulse rate, set by the photocathode laser, is variable up to1 MHz. The new machine raises the maximum beam powerfrom450Wto120 kW.Aplannedhigh-energy (HE) upgradeto 8 GeV will later triple this power.

A. Controls to limit beam loss

SLAC has multiple control systems that limit beam loss,structured in three tiers. The highest is the personnel

*Corresponding author.afisher@slac.stanford.edu

Published by the American Physical Society under the terms ofthe Creative Commons Attribution 4.0 International license.Further distribution of this work must maintain attribution tothe author(s) and the published article’s title, journal citation,and DOI.

PHYSICAL REVIEW ACCELERATORS AND BEAMS 23, 082802 (2020)

2469-9888=20=23(8)=082802(16) 082802-1 Published by the American Physical Society

protection system (PPS), which is a radiation safety systemthat interlocks access to the accelerator housing and stopsthe beam if radiation is found in occupied areas. PPSinstrumentation will not be further discussed here.The next tier is the beam containment system (BCS),

which is also a radiation safety system and constitutes themain focus of this work. The BCS “trips” (stops) the beam ifradiation from the beam or the loss of beam current indicatespossible harm to people or to safety devices like protectioncollimators. It requires robust and simple signal processing,with no knowledge of bunch timing. As safety systems, bothPPS and BCS have rigorous configuration control.The machine protection system (MPS) is allowed greater

flexibility to avoid damage from losses. It can trip the beamor reduce its repetition rate if losses exceed a threshold.Insertion of an obstacle such as a valve requires a trip, butinsertion of a screen to view the beam requires only a lowbeam rate. An excessive temperature may be addressedwith a rate limit as well. These limits can be imposed bycontrolling the rate of the photocathode laser. The MPS canselectively block one beam path (such as the beam line forone undulator) while permitting the beam elsewhere byremoving the permit for the kicker for that beam line. These

remedies can be easily reversed after adjustments are made.Since recovery from an MPS trip is much faster than from aBCS trip, MPS thresholds are set typically 10 times belowBCS, so that the MPS trips before the BCS thresholds areexceeded.

B. Beam-loss detection

Both the BCS and MPS detect losses by placingradiation sensors at points of concern along the beam path,often with some redundancy. To trip on losses from bothphotocurrent and dark current, at any beam rate, the chargecollected by a loss detector is passively integrated on acapacitor. Once this capacitor voltage crosses a threshold, atrip command is issued. The choice and location of thesesensors will be discussed in detail below.In previous installations, the BCS and MPS have

separate detectors and signal processing. To avoid thisduplication, the new beam-loss detectors will serve threepurposes: BCS, MPS, and beam diagnostics. Losses belowthe trip thresholds will provide diagnostic information tooperators for tuning the machine and locating high-losspoints to avoid or recover from a rate limit or trip. Detector

FIG. 1. Layout of the LCLS-II project, showing the new superconducting linac at left. Its beam passes over FACET-II and the originalLCLS, both normal-conducting copper linacs. Bunches from either LCLS source can be kicked into either of the two new undulators, forhard and soft x rays.

TABLE I. Parameters of the LCLS NC and SC linacs, as well as the planned HE upgrade of the SC linac.

Parameter LCLS LCLS-II LCLS-II HE

Electron energy (GeV) ≤15 4 8Bunch charge (pC) 20–300 20–300 20–300Allowed beam power (kW) 0.45 120 360Gun frequency (MHz) 2856 185.7 185.7Linac frequency (MHz) 2856 1300 1300rf pulse rate (Hz) 120 Continuous ContinuousPhotoelectron bunch rate 1–120 Hz 1 Hz to 929 kHz 1 Hz to 929 kHzPhoton energy (keV) 0.2–12 0.2–5 ≤20

ALAN S. FISHER et al. PHYS. REV. ACCEL. BEAMS 23, 082802 (2020)

082802-2

signals will be split outside the tunnel for independentprocessing for these three applications.

C. Point and long beam-loss monitors

Two detector types measure beam losses. Point beam-loss monitors (PBLMs) set limits at critical loss points,such as collimators, stoppers, and dumps. Each halocollimator includes a PBLM connected to the MPS. Amodest level of halo loss is expected, but excessive powermay result from incorrect steering of the beam or position-ing of a collimator jaw. Other PBLMs measure radiationfrom protection collimators, for the BCS. These devicesnormally do not receive a beam. In many locations down-stream of the linac tunnel, if a beam passes through the thinbeam pipe wall, a protection collimator—a barrier typically50 cm square and four radiation lengths (73 mm of steel)thick, surrounding the beam pipe—spreads out the beam toprevent a direct hit downstream by a thin beam, especiallyin certain zones with weak shielding.Long beam-loss monitors (LBLMs) detect losses over

tens to hundreds of meters and can protect the entire beampath, from the electron gun to the beam dump. LBLMs candetect losses at unanticipated spots, where no PBLM wasplaced, as well as distributed losses. We will see that anLBLM can locate photocurrent loss points within a fewmeters by the signal’s arrival time.

D. Loss thresholds and response times

The substantial increase in beam power with the SC linacis accompanied by a proportionate rise in the risk of beamloss that can endanger personnel or damage the machine,especially since losses can continue indefinitely in linacs.The loss can arrive in a fast burst as the full photocurrentbeam is lost, or a steady but smaller loss may arrive over alonger integration time. The repetition rate of the losses canbe at any beam rate from 10 Hz to 1 MHz (but a full loss at1 Hz is permitted for tuning).Field emission (dark current) from the gun or linac may

generate loss in every rf period (Table I) and would appearas a dc signal in the loss detectors. Gun dark current maybegin somewhat off orbit, travel through the entire linac,

and damage the undulator. Cavity dark current can gainsignificant energy as it travels through several cryomodulesin either direction, depending on the rf phase at emission.Safety systems for the NC linac have 8.3 ms (a 120-Hz

period) between pulses to respond to a loss exceeding anMPS or BCS threshold. The SC linac, running at up to1MHz, demands greater speed, both in detecting losses andin responding by halting the beam or limiting its rate.Losses exceeding the MPS or BCS trip thresholds must bedetected and mitigated within 200 μs so that a direct impactby the beam does not lead to melting or damaging thermalstress [4]. TheMPS canmitigate by lowering the rate, but theBCS must shut the beam off. Losses that accumulate overtimes longer than 500 ms without crossing the threshold arenot considered sufficiently harmful to warrant a trip.Table II lists the criteria for typical locations. Trip

specifications vary with the tunnel depth and other shield-ing and are given in joules of beam loss within the 500-msintegration time. BCS thresholds range from 17.5 to 500 J.MPS thresholds are generally set 10 times lower.

II. IONIZATION DETECTORS

Two categories of detector monitor beam losses for theBCS and MPS, providing different coverage and someredundancy. PBLMs are placed at likely loss locations suchas collimators. LBLMs detect losses over longer distances,at least tens of meters.At SLAC, these functions have long been based on two

types of ionization detector. After describing them, wediscuss how slow ion collection would limit their perfor-mance at high rates.

A. Protection ion chambers

A protection ionization chamber (PIC) is a PBLM placednear a component such as a collimator, to protect it fromdamage from a beam strike. Figure 2 sketches the design, astainless-steel cylinder containing a stack of 32 metal platesbiased alternately at ground or (typically) −300 V. At thisvoltage, collection of electrons from showers of ionizingradiation takes 2 μs, but ions require 1 ms of drift time.

TABLE II. Thresholds and times for the BCS and MPS faults at various locations in LCLS-II.

Trip threshold (J)Location Integration time (ms) BCS MPS

Point beam-loss monitorsProtection and halo collimators 500 50 5Beam stops and dumps 500 50 5

Long beam-loss monitorsSuperconducting linac and BSY 500 500 50Beam-transfer hall (BSY to undulator): fence open 500 2.5 1.25Beam-transfer hall (BSY to undulator): fence closed 500 17.5 1.75Beam-dump hall 500 100 10

BEAM-LOSS DETECTION FOR THE HIGH-RATE … PHYS. REV. ACCEL. BEAMS 23, 082802 (2020)

082802-3

B. Long ionization chambers

A long ionization chamber (LION) [5] detects lossesalong an extended region of the tunnel. A LION (Fig. 3) is atype of Heliax® coaxial cable with a nominal diameterof 1⅝ inch (41 mm). The corrugated Heliax outer con-ductor is made with continuous helically wound copperstrips. A copper tube forms the inner conductor. The typeused for the LION is hollow, with air or another gas servingas the dielectric. Both PICs and LIONs generally operate ata pressure of 125–150 kPa using a mixture of 95% Ar and5% CO2. LIONs are typically 30–100 m long and coverselected regions but not the full machine.A loss shower passes through the cable, ionizing the gas.

A typical bias of þ250 V on the center conductor drivescharge radially, with a collection time of 6 μs for electronsbut 6 ms for ions. The charge forms a voltage pulse movingin both directions along the cable. One end leads to a high-voltage supply, an integrator, and a digitizer. The other endhas a high-impedance termination drawing a small dccurrent, called the “pedestal,” which indicates that thesystem is functioning. However, it masks detection ofsmaller loss currents.

C. 1D model of an ionization chamber

The possible loss of significant beam power and an ioncollection time far longer than the interval between pulsesboth suggest that ion space charge may accumulate insidean ionization chamber. An experimental and numericalstudy of ionization chambers for protons in the NuMIexperiment at Fermilab [6] found that screening by positiveions can completely null the field near the positive

electrode(s), creating a “dead zone.” This motivated thedevelopment of a self-consistent one-dimensional model ofcharge flow and field evolution in the SLAC devices.The model assumes a uniform ionization rate IðtÞ per

unit volume generated by beam loss and a recombinationrate per unit volume βnine proportional to the product ofelectron and ion densities. The PIC behavior is computedfor the x coordinate between a pair of plates. The signalcurrent is then scaled by 31 to account for all plates. TheLION model computes the radial behavior per unit lengthin a uniformly ionized cable.This numerical model (as well as the experiments dis-

cussed later) uses pure argon at a pressure of 100 kPa. Argon,a common gas in ionization chambers, avoids both thecomplexity of molecular fragmentation and also the creationof negative ions, which move as slowly as positive ions andsowould introduce additional space-charge accumulation, inthis case near the negative electrode(s).At this pressure, ions and electrons have mean free paths

much smaller than the chamber size, and so particle drift ismodeled by an ion mobility μi (with a velocity vi propor-tional to the electric field E) and an electron mobility μe(which retains a weak E dependence). The ion and electronparticle fluxes Ji;e also include a diffusive component withcoefficientsDi;e:Gauss’s law givesE self-consistently. Thegoverning equations in SI units are

Ji;e ¼ �μi;eni;eE −Di;e∇ni;e;

∂ni;e∂t ¼ −∇ · Ji;e þ I − βnine;

ε∇ ·E ¼ eðni − neÞ: ð1ÞHere, e is the elementary charge. The effective dielectric

ε is ε0 in a PIC but slightly higher in a LION due to thecable’s construction.

FIG. 2. Cutaway view of a PIC. The housing is a stainless-steelcylinder with a 114-mm outer diameter and a 270-mm length.The 32 plates each have an area of 56.5 cm2 and ared ¼ 6.15 mm apart. The plates are alternately grounded orconnected to a bias voltage.

FIG. 3. Sketch of a LION. The nominal inner- and outer-conductor diameters are 18.1 and 46.5 mm, respectively.

FIG. 4. Model of the radiation field showing a loss on aprotection collimator at z ¼ 0. The color scale gives the dose ineV=g per incident 4-GeV electron (for which 1 eV=g ¼250 nGy=J). Six PICs (with axes perpendicular to the page) areinserted at z ¼ 50, 100, and 200 cm and at r ¼ 50 and 100 cm.Shielding by their steel housings creates the shadows. The dose atthese PICs peaks at 170 and 36 μGy=J at r ¼ 50 and 100 cm,respectively, and drops slowly with increasing z.

ALAN S. FISHER et al. PHYS. REV. ACCEL. BEAMS 23, 082802 (2020)

082802-4

Electrodes are charge sinks but not sources: Electrons orions can flow in but not out. The surface charge Q on eachelectrode is determined self-consistently as an integral overthe electrode’s surface S. In addition, the bias supplyprovides the charge required to maintain a constant voltageV0, given by a line integral du between any points a and bon the two electrodes:

Q ¼ ε

ZSE · dS;

V0 ¼ −Z

b

aE · du: ð2Þ

A current Iext flows through the external circuit from thebias supply. The current IS changing the surface charge isthe sum of Iext and the collection of electrons or ions:

IS ¼ Iext � eZSJi;e · dS: ð3Þ

D. Loss on a protection collimator

To complete the model, consider a typical loss point, aprotection collimator. These are mostly found in thebeam-transfer hall (BTH), between the beam switchyard(BSY) and the undulator hall. Because of the hilly terrain,the BTH lies at ground level and is shielded within aconcrete bunker designed for the far lower beam power ofthe NC linac. Additional measures, including low BCSand MPS thresholds, are needed for protection from theSC linac beam.Because a high-power beam loss could expose people

outside to a harmful radiation dose, the region adjacent tothe walls is restricted by an interlocked fence. For furthersafety, any beam escaping the beam pipe must hit a nearbyprotection collimator surrounding the beam pipe furtherdownstream. Scattering in the collimator’s 73-mm thick-ness of stainless steel spreads out the radiation shower.

FIG. 5. Response to a 120-kW loss of a 1-MHz beam on a protection collimator over 2 ms. (a)–(c) Response of a PIC placed atr ¼ 0.5 m and at the radiation peak in z in Fig. 4, as a function of time and, in (a),(b), of position between adjacent plates. (d)–(f)Response of a LION at r ¼ 1 m and spanning the radiation peak, as a function of time and, in (d),(e), of the position between the innerand outer coaxial conductors. (a),(d) Electric field. (b),(e) Electron density. (c),(f) Ion and electron currents collected at the cathode andanode, respectively. Because the electron current fluctuates over a wide band during the 1-MHz period, the average current in eachperiod is also shown.

BEAM-LOSS DETECTION FOR THE HIGH-RATE … PHYS. REV. ACCEL. BEAMS 23, 082802 (2020)

082802-5

A hollow layer within the collimator is filled with nitrogenabove atmospheric pressure. If a beam burns through thesteel to this gas pocket, a pressure switch shuts thephotocathode laser off.The FLUKA model in Fig. 4, showing the dose from beam

loss at a protection collimator, relates the lost beam energyto the radiation dose nearby [7], as a function of thedistance radially (r) and longitudinally (z) from thecollimator at r ¼ z ¼ 0. By definition, a dose of 1 Gydeposits 1 J=kg, and an energy of 27 eV is needed to createan electron-ion pair in argon. We can then calculate theionization in a PIC or LION near the collimator.

E. Model results for PICs and LIONs

Figures 5(a)–5(c) show the response of a PIC to a lossof 120 kW—the allowed power in the superconductingbeam—on a protection collimator for the first 2 ms after theloss begins. The PIC is placed at r ¼ 50 cm and at the zcorresponding to the peak of the radiation field in Fig. 4,170 μGy=J. The BCS loss limit for a protection collimator,50 J (Table II), results in a dose of 8.5 mGy.The electric field is shown in Fig. 5(a) as a function of

the time and position between adjacent plates. The groundis at x ¼ 0, and the standard bias of −300 V is applied tothe plate at x ¼ d. The field between the parallel plates isinitially uniform, and all the field lines terminate on surfacecharge on the plates. As the ions accumulate faster thanthey can travel to the negative plate, some field lines fromthe negative plate terminate on ion space charge beforereaching the positive plate; after 160 μs, the field near thepositive plate is completely screened. We call the resultingregion of zero field the “dead zone.” Since the bias supplymaintains a constant voltage difference, the field must growtoward the negative plate.Transport of electrons and ions (except for diffusion)

halts in the dead zone, and both the electron [Fig. 5(b)] andion densities grow quickly there with the periodic (1 MHz)loss. The collected electron current [Fig. 5(c)] shows ahighly nonlinear drop. Note that this nonlinearity occursafter 160 μs, a time corresponding to a beam loss of 19 J,less than half of the BCS limit.A LION receives a dose near the peak value over a length

of about 1 m, and so we use 35 μGy=J for r ¼ 1 m. InFigs. 5(d)–5(f), the LION exhibits behavior similar to thatof the PIC, exacerbated by a larger electrode separation.Again, a dead zone forms, and the electron density risesrapidly there. The collected electron current becomesnonlinear after 125 μs, corresponding again to 15 J ofloss. The bias is the standard þ250 V, with the ground onthe outer conductor. A negative bias would be worse: The1=r dependence of the coaxial field makes it weaker at theouter electrode and so reduces ion transport in this criticalregion, making the field easier to screen.We see that both ionization detectors can become

nonlinear at loss levels below the BCS thresholds.

Such behavior is inappropriate for a critical safety system.Although the MPS is designed to trip at a lower losses,the BCS must work independently. Some improvementcould be obtained with greater distance and higher voltagebut would not fully address these failings. Also, the gas-distribution system now deployed over the acceleratorcomplex has limitations in reliability, and extending itwould be expensive. These considerations led us to inves-tigate other detector types.

III. LBLMS FOR LCLS-II

A. Optical fibers

In place of LIONs, SLAC has selected optical fibers asLBLMs for the LCLS-II project. A loss shower passingthrough a fiber emits Cherenkov light, a portion of which iscaptured in a fibermode and carried to a photomultiplier tube(PMT) at one end. For sensitivity, each fiber is relatively thick—with diameters of 600, 660, and 710 μm for the core,cladding, and buffer, respectively—and is encased in a 2-mmjacket of black polyurethane, for protection and opacity. Thefiber, type FBP-600660710 from the Polymicro division ofMolex, uses a radiation-resistant quartz for both core andcladding. This and related types were subjected to extensivetests of radiation hardness—up to 12.5 MGy—for use in theend cap of the CMS detector on the LHC at CERN [8–10].The FBP type was found to be superior, especially for redwavelengths around 700 nm.A distance of 4 kmmust be protected, from the gun to the

beam dump, in most places using two fibers both forredundancy and also for different viewing angles. Thisdistance is subdivided into lengths of roughly 200 m.Variations occur to span functional sections, such as theinjector or the L2 linac between the first and second bunchcompressors.Dividing the fiber in this way offers two benefits. First,

attenuation of light in the fiber limits the maximumsegment length. Also, by restricting the length and sam-pling the PMT waveform at a high rate, we gain twodiagnostic capabilities. First, a loss point can be localizedwithin 3 m by measuring the arrival time of the loss signalfrom a photocurrent bunch at the PMT, without ambiguityfrom pulse pileup. When the PMT is placed at the down-stream end of the fiber, for reasons addressed later, lossesalong a 200-m fiber from one bunch span 330 ns, wellunder the minimum bunch spacing of 1 μs in the SC linac.Second, when a wire scanner measures the transverse

bunchprofile, software extracts theportion of the PMTsignaldue to beam current intercepted by the wire and correlates itwith thewire position to determine the beam size, without theneed for a dedicated detector at each wire scanner.

B. Fiber attenuation and PMT selection

If light is strongly attenuated in the fiber, then two equallosses, one occurring at the end of the fiber near the PMT

ALAN S. FISHER et al. PHYS. REV. ACCEL. BEAMS 23, 082802 (2020)

082802-6

and one at the distant end, will generate PMT signals ofvery different amplitudes. If trip thresholds are set based onthe stronger loss, unacceptable losses at the distant end maynot trip the beam. If the thresholds are based on theattenuated signal, then tolerable losses at the near endmay cause unnecessary trips. Our criterion is that thesesignals must not differ by more than 3 dB (a factor of 2).Then the lowest curve in Fig. 6, which plots the inherentattenuation of a 200-m fiber, restricts the wavelengths toλ ≥ 510 nm; shorter wavelengths have too much loss.A longer cutoff is needed to account for the increase in

attenuation after irradiation over several years of use.Measurements made for CMS provide a fit to radiation-induced attenuation as a function of the dose and wave-length [9] for a similar Polymicro quartz fiber, type FVP.Its attenuation is somewhat higher than that of FBP afterextensive neutron and gamma irradiation [10], and so the fit[9] allows a conservative calculation of attenuation in ourFBP fiber as a function of the dose.It is unlikely that the entire fiberwouldbe subjected to high

levels of radiation. Instead, Fig. 6 presents a scenario inwhich the fiber passes a strong loss point, where 10 of its200m receive various high doses of radiation over time.Near700 nm, attenuation remains largely independent of the doseand easily satisfies the 3-dB criterion. Wavelengths can berestricted to this range by covering the PMTwindow with along-wave-pass filter with a sharp cutoff. The quantumefficiency (QE) of the PMT photocathode itself blocks evenlonger wavelengths. The combination effectively acts as abandpass filter with sharp edges.We must, of course, allow for the reduction in the total

signal due to this rejection of shorter wavelengths, espe-cially since Cherenkov emission, although broadband, hasan intensity dI=dxdω that is proportional to ω [11] and soappears blue. However, a PMT responds not to the intensitybut to the photon flux dN=dxdω, which is flat with

frequency (although, expressed in terms of wavelength,dN=dxdλ ∼ λ−2). We calculate the relative PMT signal byan integration in wavelength over the product of theCherenkov emission, the fiber transmission, the color-filtertransmission, and the QE of the PMT. Our measurementsusing the NC linac (presented below) confirm a satisfactorysignal in this spectral range near 700 nm.In an area such as the BTH, the beam must trip at a lower

radiation level. If the wavelengths are restricted to theregion near 700 nm, the trip threshold would be set at acorrespondingly low signal, which could be sensitive tonoise. However, since exposure limits require a lowradiation dose there, we can afford to raise the signalstrength by extending the cutoff wavelength toward thegreen (but remaining above the zero-dose wavelength limitof 510 nm). A fiber dose limit of 1 kGy in this zone allowsseveral years of use, and calculations based on Fig. 6 findthat a cutoff of 555 nm satisfies the 3-dB criterion. Incontrast, a fiber in the linac, where the threshold is higher, isalso expected to receive a higher dose over a similar time.There, the filter cutoff is 675 nm, allowing a dose of up to100 kGy.To verify that a fiber’s attenuation has not increased to an

unacceptable level, the transmission of each fiber will becontinuously monitored with the “heartbeat” introduced inSec. III C below.Few photocathode materials are sensitive at wavelengths

near 700 nm. The strongest response comes from GaAsP,used in the Hamamatsu H7422P-40 PMT, which has a QEof 30% at this wavelength. However, the low work functionneeded in any red-sensitive photocathode increases itsdark current. This background would mask the detectionof structure dark current from an emission site in theelectron gun or superconducting rf cavities. Once formed,such a site emits in each rf period, leading to a quasi-cwsource of potentially damaging radiation. We calculate thatthe anode dark current of the PMT should be limited to1 nA, low for red-sensitive cathodes. However, PMT darkcurrent is thermal, dropping by about a factor of 2 for every5 °C of cooling until this benefit levels off at 0 °C [12]. Thesecond advantage of the H7422P-40 is the built-in Peltiercooler that maintains an operating temperature of 0 °C.

C. Fiber layout

Both ends of the fiber up come up from the tunnel to theklystron gallery, about 11 m above the tunnel floor.Figure 7 illustrates the layout. One end runs to a rackwith an LBLM chassis containing the PMT, with itsentrance window covered by a suitable optical filter andan SMA-905 fiber connector.Figure 7 shows a similar chassis at the other end of the

fiber, where it connects to an SMA-905 containing a light-emitting diode (LED) emitting at a wavelength just withinthe edge of the filter’s passband. The LED provides acontinuous self-check of the fiber system, called the

FIG. 6. Radiation-induced attenuation in a 200-m fiber irradi-ated over 10 m, for various doses in kGy.

BEAM-LOSS DETECTION FOR THE HIGH-RATE … PHYS. REV. ACCEL. BEAMS 23, 082802 (2020)

082802-7

heartbeat. Its failure trips the BCS. Details are discussed inSec. V C below.If a fiber needs replacement after several years, it can be

done quickly, without entering the tunnel and withouthalting the beam. “Ducts” of polyethylene tubing with a6-mm inner diameter run along the fiber routes. A specialtool, aided by the flow of nitrogen at 700 to 1000 kPa,pushes a fiber through a duct. In our tests, a 200-m fibertravels from the rack at one end to the rack at the other endwithin a few minutes. The fiber is ready for use afterSMA connectors are attached to the ends and the tips arepolished.

D. Modeling the fibers

The geometry for capturing Cherenkov light in a fiber iscomplex. This quartz fiber captures only rays travelingwithin an angle of 8.7° to the axis, allowing them to travelto the PMT. The axis of the Cherenkov cone emitted by arelativistic particle from a loss shower passing through thefiber core lies at 46° to the direction of travel, due to therefractive index of the medium. Only part of the cone canfall within the capture angle. The most effective alignmentoccurs where the particle’s entering angle is close to 45° tothe axis, which places the onset just downstream of the losssource. However, showering and screening by interveningobjects can alter the signal.This process has been modeled with a Monte Carlo

extension to FLUKA tracing the Cherenkov emission [13].This simulation finds that 90% of the signal is integratedwithin 20 m downstream of a thick target (such as a beamstop) and within 50 m from a target of medium thickness,such as a protection collimator. Upstream contributions areminimal. Because the thick target absorbs some of the lossshower, its signal is lower (Fig. 8). The transverse distanced of the fiber from the beam line has a soft effect on thesignal, falling off as d−0.79 (for d in centimeters).

IV. PBLMS FOR LCLS-II

A. Diamond detectors

In place of PICs, the LCLS-II project has selecteddiamond-based detectors as PBLMs. As a loss showerpasses through a thin diamond chip, every 13 eV of lostenergy creates an electron-hole pair. A bias voltagebetween metalized electrodes plated on the opposite facesof the chip separates the pair. The diamond resembles asolid-state ionization chamber but with the importantdifference that both electrons and holes take only a fewnanoseconds to reach the electrodes: No pileup happens atthe 1-MHz beam rate. Unlike an LBLM, loss localizationthrough arrival time is not important for a PBLM, sinceeach detector is in a fixed location near expected losspoints.We investigated three types of diamond detectors, all

originally developed for CERN and made by Cividec [14]:the single-crystal B1, the polycrystalline B2, and the high-radiation polycrystalline B4. Both the B1 and B2 modelshave a wide dynamic range, responding to the passage of1–106 minimum ionizing particles (from a loss shower thathas cascaded to the minimum of dE=dx). The B2 sensorhas the benefit of an active volume that is four times largerthan the B1. The B4 is smallest and is designed for a lowersensitivity to avoid saturation in a high-radiation environ-ment. The experiment presented in Sec. VI C found thatsignals from both polycrystalline types exhibit a slow decaywhen the loss stopped. The single-crystal B1 shows no suchtail and so was chosen for the high repetition rate of theSC linac.The continuous heartbeat test for the LBLMs was

introduced earlier. A similar test was devised for thediamond detectors, accomplished by modulating the biasvoltage. Both are detailed in Sec. V C.

FIG. 7. Typical fiber layout in the linac tunnel, including thePMT at the downstream end and the heartbeat LED at theupstream end. The splitter that provides a signal for threefunctions is indicated. A solid line indicates the fiber pathbetween the two penetrations shown; the two fibers runningbeyond this region are shown with dashes. FIG. 8. Simulated signal (nA=W) integrated along the fiber as a

function of the horizontal distance to the loss point. The uppertrace shows loss on a target of medium thickness (e.g., aprotection collimator); the lower traces shows a thick target.

ALAN S. FISHER et al. PHYS. REV. ACCEL. BEAMS 23, 082802 (2020)

082802-8

B. Modeling the diamonds

Cividec has simulated electron-hole pair productionin single-crystal diamond by high-energy electrons, posi-trons, protons, neutrons, and photons [15]. These resultswere incorporated into another FLUKA extension [13] thatuses the fluences and energies of the various showerparticles in a Monte Carlo calculation of the signal froma B1 diamond. Note that the detector’s signal depends on itsvolume (4 × 4 × 0.5 mm3 for the B1) but not on itsorientation.The simulations show that this diamond can detect losses

on a thick target above 100 mWat a distance of 1 m, belowthe requirements in Table II. However, the response falls offquickly with distance from such targets. This is an issuewith any loss detection in a self-shielding target.For thinner targets, Fig. 9 shows that the signal from a

loss on a protection collimator is nearly constant over alongitudinal distance of as much as 40 m, when measuredat a transverse distance of 1.5–2 m from the beam pipe. Wesee also that the signal from glancing incidence on a beampipe exceeds that of the protection collimator for distancesbelow 20 m. All the protection collimators in an area canthen be covered by locating one diamond every 20 mbetween beam lines. With many such collimators onadjacent beam lines in the BSY and BTH, this schemeoffers a considerable economy in the number of detectors.In a test of this effect in end station A at SLAC, a beam

was incident on a 16-cm-square, 80-mm-thick iron block.A B1 diamond was placed 166 cm to the side, suspendedfrom the shielding wall on a motorized clothesline, andmoved downstream of the block from −1 to þ17 m. Thesignal was not flat with distance, as in Fig. 9, but peaked7 m from the block and dropped by half at 12 m. A more

detailed analysis, accounting for the smaller width of theblock and screening by obstacles that could not be removedfrom the tunnel, found consistency with the model.

C. Scintillators

The cw electron gun for the SC linac has an outputenergy of 750 keV, low compared to 5 MeV from thepulsed gun of the NC linac. Early commissioning of thegun and the first 3 m of beam line took place over severalmonths in 2019 and again for 3 months in 2020, duringinstallation of linac cryomodules. A temporary Mo cathodewas used during initial tests, bakeout, rf processing, andinitial observations of dark current. A Cs2Te photocathodewas then installed for photocurrent commissioning.A short fiber alongwith two diamonds, one on the exit face

of the gun and the other 1 m downstream by the entrancewindow for photocathode laser light, were installed forinjector commissioning. Before electrons accelerate in thefirst cryomodule, beam loss can generate only a weakresponse from these detectors, especially the fiber due tothe low-energy cutoff of Cherenkov radiation. Consequently,two scintillators of cerium-doped yttrium aluminum perov-skite (YAP:Ce, from Crytur) with PMTs were installed nextto the diamonds. Dosimeters placed beside these detectorsmeasured the total dose in the 2020 test.The scintillators easily detected dark current from the

Mo photocathode. Adjusting the focusing solenoid at thegun output shifted the loss from one scintillator to the other.During part of the 2020 run, when the gun rf ran for about 4h per day with dark current only, the second scintillatorsaturated the electronics (∼10 V, integrated as in Fig. 10but with a larger C1). When the rf shut off, the scintillatorexhibited a large offset (∼2 V) that decayed slowly withexponential components of 8 and 100 h. The first scintil-lator reached 5 V with a much smaller offset (∼5 mV) butwith the same decay. As expected, the diamond signalswere much smaller (a few millivolts) but shut off immedi-ately without offsets or tails. Radiation levels were notunusual for an accelerator: The dosimeters recorded 17 Gyat the second location but only 140 μGy at the first. Becauseof these tails, the scintillators are not suitable as PBLMs,but the diamonds should prove satisfactory.

FIG. 9. Simulated current (pA) from a B1 diamond placed 2 mtransversely from the beam pipe, as a function of the longitudinaldistance from a 1-W beam loss on a protection collimator (redline) or from glancing incidence on a 1.5-mm-thick stainless-steelbeam pipe.

FIG. 10. Front-end circuit showing the input integrator andsplitter. The diagnostics waveform is used only with the fiber.

BEAM-LOSS DETECTION FOR THE HIGH-RATE … PHYS. REV. ACCEL. BEAMS 23, 082802 (2020)

082802-9

V. SIGNAL PROCESSING

A. Unified design

The requirement for beam-loss detectors suitable forhigh loss power at high beam rates demands a rapid shutoff,with a time budget of 200 μs from first exceeding the tripthreshold to mitigation (by stopping the beam or, for theMPS, possibly lowering the rate). The electronic latencymust occupy a small fraction of this interval by respondingto a threshold breach within 10 μs. Consequently, a newdesign was developed for the signal-processing electronicsand the overall BCS architecture [16].Signals from both the LBLMs and PBLMs are integrated

over 500 ms. The choice of a negative bias for the diamondsmatches their output signal polarity to that of the PMTs.Similar integrated voltage levels can be obtained by selectingsuitable components for the charge integrator. Furthermore,both sensors perform a continuous heartbeat self-check(Sec. V C) based on a lock-in algorithm in a digital signalprocessor (DSP). These commonalities enable a unified frontend for the electronics of both sensor types offering signifi-cant savings in engineering and design verification.

B. Integrator and splitter

The charge from the diamond sensor or PMT is split intolow- and high-frequency components with the passivecircuit in Fig. 10. Both sensors are current sources,generating a charge proportional to the loss. The chargeis transferred immediately upon arrival to the integrationcapacitor C1, which accumulates the charge from a train ofloss pulses. A low-leakage film capacitor is used, with avalue based on the expected charge at the BCS lossthreshold. A capacitance of 33 nF is used for thePBLMs; the LBLMs use 330 nF to 1 μF, depending onthe location. For diamonds, C1 includes the capacitance of

the cable from the tunnel. Because each fiber ends at achassis containing a PMT next to the integration circuit,cable capacitance does not enter into the component valuesthat set the integration time constant.The capacitor slowly discharges through resistors R1, R2,

and R3 in series. A 500-ms time constant is set by R1, whichranges from 0.5 to 1.5 MΩ for the LBLMs to 15 MΩ forthe PBLMs. The integrator output VL is buffered and scaledby an amplifier with a gain of 2–50, set by switches on theboard. The resulting level is compared to the BCS lossthreshold. The amplifier needs a low input bias current, toavoid leaking charge from the capacitor, and a low inputoffset voltage, to avoid affecting the signal on the com-parator. Because the klystron gallery is not temperaturecontrolled and the racks are not cooled, the temperature ofthe electronics can vary diurnally and seasonally from 0 to50 °C; consequently, the selected amplifier has a lowtemperature drift to maintain a low offset. The amplifiedcapacitor voltage is buffered by a second amplifier withunity gain before transmission to the MPS digitizer chassis(which in many cases is in another rack).Each detector has a different threshold, calculated for

various loss scenarios. To avoid unintentional changes overthe network, the threshold is set locally, with thumbwheelswitches on the front panel. Access to this unit is protectedwith a lock. The threshold level is continuously monitoredby a programmable logic controller (PLC), which halts thebeam if the value read does not match the expected level.The high-frequency component VH of the detector

output is terminated in 50 Ω by R2. For the LBLM, VHis buffered by a fast amplifier and sent to a 350 MSample=sdigitizer. The output waveform provides a diagnostic tolocate the beam loss. The arrival time of the loss peak at thedigitizer allows us to determine the source of a loss withinabout 3 m. During a wire scan (Sec. VI E), software extracts

FIG. 11. A functional block diagram of the BLM electronics. Highlighted boxes show the direct path for beam loss.

ALAN S. FISHER et al. PHYS. REV. ACCEL. BEAMS 23, 082802 (2020)

082802-10

the portion of the waveform showing the signal from thebeam’s passage through the wire and correlates it with thewire position to measure the bunch profile projected ontothe direction of the wire’s motion. The software providesthree sampling windows, configurable in width and delay,to numerically integrate loss peaks and subtract pedestal.Figure 11 shows the various branches and the overall

processing scheme. However, it was noted earlier that thePBLM has no need for a fast waveform.

C. Self-checking

We encode the LBLM self-check heartbeat signal bydriving a small 0.8-Hz sinusoidal current in the LED at theupstream end of the fiber. A DSP running an algorithm likethat of a digital lock-in amplifier sensitively detects thismodulation in the PMT output. A missing heartbeat—indicating a failure of the fiber, the PMT, its power supply,or the electronics—trips the BCS. Modulating the currentrather than the voltage was found to be more tolerant toLED temperature variations. The control system archivesthe detected heartbeat amplitude, so that any gradualdegradation of the fiber transmission can be tracked.To create a PBLM heartbeat, its −250-V bias is

modulated by 5 V peak to peak. The resulting smallcurrent through the diamond capacitance is also detectedby a lock-in implemented on a DSP. The modulation variesthe few-nanosecond transit time to the electrodes propor-tionately but with no effect on the total charge collected.The modulation frequency of 0.8 Hz, used with both

detector types, avoids any subharmonic of a beam rate,which might influence the demodulated amplitude. Also,this low frequency passes through the RC integration filter.The narrow-band filter in the DSP is slow, requiring over aminute to settle after the power comes on and a similar timeto trip if the signal is lost, but this is sufficiently fast todetect a rare failure of a BCS component (and far betterthan the weekly manual testing used previously).For the PBLM, the same microcontroller modulates the

drive signal and detects the modulation; thus, the phase ofself-checking tone is known and stable. For the LBLM, themodulation and detection are far apart, at the two ends ofthe optical fiber. The modulation and demodulationfrequencies are matched by counting in firmware 75 cyclesof the 60-Hz ac power line (60=75 ¼ 0.8).

D. LBLM temperature and humidity

The integrated Peltier cooler in the Hamamatsu modulecan maintain a PMT temperature of 0 °C, to lower the darkcurrent, only if the ambient temperature does not exceed35 °C. On a hot summer afternoon, the temperature inside arack in the klystron gallery of the SLAC linac may reach50 °C. Consequently, we added an external thermoelectriccooler that transfers heat directly from the PMT housing.To avoid condensation onto the PMT window, the

external cooler is configured to operate only when the

temperature of the PMT housing reaches 30 °C, which isabove the dew point expected in the gallery. The chassiswas tested in an environmental oven with a maximumtemperature of 50 °C and a relative humidity as high as100%. No condensation was observed on the PMTwindowor housing. The detected self-checking amplitude remainedstable, and the input voltage offset (from the PMT darkcurrent and electronics) remained within our requirements.

E. PBLM cables and connectors

Unlike the fiber and PMT, a diamond detector requires abias cable to carry the high voltage from the rack outsidethe tunnel to the detector inside and a signal cable to carrythe output current from the detector back to the rack. Thesignal current at the MPS threshold is small, typically a fewnanoamps, and must not be shunted through the leakageresistance of the coaxial dielectric before arriving at therack. Hence, the 15-MΩ discharge resistor in the passiveRC filter in Fig. 10 must be well below the leakageresistance of 50–150 m of cable, a specification thatmanufacturers do not provide (and often do not know).Consequently, we tested long cables of different types forleakage resistance, noise immunity (good shielding), andattenuation. Double-shielded cable was preferred to avoidnoise pickup from the various kicker magnets that switchthe high-rate electron bunches to their various destinations.To find the leakage resistance, each cable was charged to

500 V (which kept within its voltage specification). Then, apicoammeter with a 10-fA resolution measured the currentthrough the dielectric. In all cases, the resistance was foundto be at least 200 MΩ. The cable was then disconnected anddischarged through a 50-Ω termination. Several cable typesexhibited a peculiar and quite unexpected behavior: Whenwe reconnected one end of the cable to the meter and leftthe other end open, the meter registered a persistent dccurrent of as much as 1 nA. Despite discharging for as longas a few days, the current resumed once the cable wasreconnected to the meter. A voltmeter used to check thepicoammeter read a consistent value of a few millivolts.Times Microwave, a cable manufacturer, suspected thatfriction between the dielectric and the material of the outerconductor generated the charge. They sent spools of severalcable types for testing. We could not elucidate the mecha-nism but ultimately found that their SF-223 (a variant ofRG-223) does not exhibit this persistent current.Reynolds type C, a high-voltage cable widely used

across multiple systems at SLAC, was selected for thebias. To test for pickup, the diamond and both cables wereplaced inside coils of large-diameter Heliax cables driving aprototype kicker in its test stand. The cables ran to theprototype diamond electronics and to an oscilloscope. Nopickup was seen in the integrated detector signal.Triaxial cable was tested with a guard drive connected to

the inner braid, which held the average voltage drop at zeroacross the inner dielectric, to null leakage in the cable from

BEAM-LOSS DETECTION FOR THE HIGH-RATE … PHYS. REV. ACCEL. BEAMS 23, 082802 (2020)

082802-11

this braid to the center conductor. Ultimately, this wasfound unnecessary and, in fact, increased the leakage due tothe transition from the triaxial cable to the coaxial output ofthe sensor.Connectors mating with the diamond detectors are

custom variants of LEMO type 00 (for bias) and SMA(for signal). The insulators use polyether ether ketone(PEEK) rather than the standard polytetrafluoroethylene(PTFE), to significantly increase the radiation tolerance.

VI. EXPERIMENTAL TESTS

A. LION and PIC versus fiber and diamond

In the experiment depicted in Fig. 12(a), a 180-pC,5.115-GeV electron beam hit tungsten plates at 5 Hz. Theloss signals were collected by several detectors placed toone side: a PIC, a short LION, two polycrystalline diamonddetectors (a standard B2 and a high-radiation B4), and anoptical fiber.Figure 12(b) shows that both diamonds are faster and

less noisy than the PIC. The pulse from the B4 diamond isabout 5 times higher than that from the B2, although abouthalf the width. However, the B4 is 7.5 times closer. Theinverse square of the distance (an approximate correction,since the radiation field is not spherically symmetric)suggests that the B4 is less sensitive by more than a factorof 20. Ideally, an ionizing particle passing through adiamond produces 36 electron-hole pairs per micrometerof travel. The B4 is both 5 times thinner than the B2 andalso intercepts fewer particles with an electrode areathat is 9 times smaller. The more fractured crystal

structure of the B4 also lowers the efficiency of chargecollection [17].The LION signal in Fig. 12(c), processed by standard

SLAC electronics, was very slow and required a muchlonger timescale. It also exhibited significant 60-Hz modu-lation. Compare this to the fast signals in Fig. 12(d) fromdifferent photosensor types at either end of the fiber. Thefiber ran along the LION, around a delay loop shieldedfrom radiation, and back along the LION. One end had aPMT (Hamamatsu R7400U-06, a small blue-sensitivetube); the other had a silicon photomultiplier (SiPM,Hamamatsu C13360-6050CS), an array of small avalanchephotodiodes that act as a solid-state PMT by combining theoutput of all the diodes. Ultimately, we rejected the SiPMdue to its higher dark current, lower dynamic range, andcross talk between neighboring diodes. The SiPM pulsecould be shortened with a transimpedance amplifier.The upstream ends of the LION and fiber were 90° from

the tungsten target, 1.5 m away. They ran downstream for1.5 m and so ended 45° from the target in the forwarddirection. The large initial peaks for the PMT and SiPMsignals showed the capture of forward-going light from aloss shower. The second peak, delayed by the longer path inthe fiber, came from backward-going light from the showerbut was not favored by the layout, since the fiber had littlebackward extent. Next, we present an experiment makingthis forward-backward comparison with a long fiber.

B. PMT location: Upstream or downstream

Cherenkov light can couple into a fiber mode traveling ineither direction. Each end of the fiber offers advantages as

FIG. 12. (a) Layout of an experiment comparing old and new detectors as 5.1-GeV, 180-pC electron bunches hit tungsten plates.Signals from (b) a PIC and diamonds; (c) a LION, with a very slow timescale and large 60-Hz noise; and (d) a PMT and a siliconphotomultiplier (SiPM).

ALAN S. FISHER et al. PHYS. REV. ACCEL. BEAMS 23, 082802 (2020)

082802-12

the PMT location. We measured the effect with a blue-sensitive PMT (Hamamatsu R7400U-06) at each end ofa 155-m fiber. The blue PMTs also allowed us to comparethe measured and calculation attenuation, since the lossis higher over this distance (30 dB=km at 420 nm vs5 dB=km at 700 nm).Consider an electron bunch lost at z after passing the start

(z ¼ 0) of a fiber of length L at t ¼ 0. By adding the traveltimes of the electrons and the light, we find that anupstream PMT (PMT1, at z ¼ 0) receives its signal attuðzÞ ¼ ð1þ nÞz=c, where the refractive index n ≈ 1.5.The signal arrives at a downstream PMT (PMT2, at z ¼ L)at tdðzÞ ¼ ð1-nÞz=cþ Ln=c. Two losses a distance Δzapart produce signals separated by Δtu ¼ ð1þ nÞΔz=c ≈2.5Δz=c and Δtd ¼ ð1-nÞΔz=c ≈ −0.5Δz=c. The signals

arrive at PMT2 in reverse order and 5 times closer in time,giving PMT1 an advantage. Simply zooming in corrects forthis compression until reaching a resolution limit deter-mined by cable dispersion, PMT rise time, or digitizerbandwidth. In our case, loss localization to 3 m with adownstream PMT is achievable and sufficient.The counterargument to placing the PMTat the upstream

end is that the beam-loss shower is stronger in the forwarddirection and so sends more Cherenkov light to a down-stream PMT.In the test, corrector magnets steered the beam into the

beam pipe to generate losses at various distances from abeam stopper at the downstream end. PMT loss signals(Fig. 13) show changes in both the arrival time andamplitude. Figure 14 plots the signal ratio PMT1/PMT2.As the loss shifts downstream, the attenuation grows forPMT1 and drops for PMT2, doubling the relative attenu-ation. The slope of the blue line in Fig. 14 came not from afit but from a calculation integrating over the wave-length dependence of Cherenkov emission, the fiber trans-mission, and the QE of this PMT. A fit to the measuredpoints provided the offset of this line and indicated abackward-forward signal ratio of 27%. This result moti-vated the choice to put the PMT at the downstream end ofeach fiber.

C. Tails on polycrystalline diamond signals

In comparing the responses of side-by-side B1, B2, andB4 diamond detectors, we found the B2 and B4 exhibited aslow tail when the loss stopped. In Fig. 15, about 1 W ofthe LCLS beam was scraped on a collimator monitored bya B1 and a B2, as the beam rate was switched between 0and 120 Hz. The polycrystalline B2 exhibited dual timeconstants, initially 2 s followed by 18 s, due to charge trapsat domain boundaries. The B4 shows an even longer decaytail and a similarly slowed rise. The single-crystal B1 hasno tails and so was chosen by the LCLS-II project for the

FIG. 13. Signals from PMT1 (green trace, upstream end of fiber) and PMT2 (blue, downstream end); 200 ns=div and 1 V=div into50 Ω. The loss points are (a) 82 and (b) 39 m upstream of the beam stopper at the downstream end. As losses move downstream, thepeaks move rightward for PMT1 but leftward for PMT2, and the ratio of their amplitudes PMT1/PMT2 decreases. The pulse widths andpeak separations of PMT2 are compressed by 5 relative to PMT1.

FIG. 14. PMT1/PMT2 ratio. The calculated slope comes onlyfrom the properties of the Cherenkov light, fiber, and PMT. Thevertical offset, from a fit to the measurements, determined that thesignal from the backward loss shower was 27% of the forwardsignal.

BEAM-LOSS DETECTION FOR THE HIGH-RATE … PHYS. REV. ACCEL. BEAMS 23, 082802 (2020)

082802-13

high-rate beam. The mechanism driving these tails in thepolycrystalline detectors is not well understood [17].This experiment also gave the first opportunity to test the

continuous heartbeat self-check, described earlier, in whichthe bias voltage is modulated at 0.8 Hz. In Fig. 15, the B1shows a noticeable sinusoid due to a large modulation. Thefilter was subsequently improved, so that the lock-in nowrobustly detects the heartbeat from a modulation currentsmaller than any loss signal of interest.

D. Field emission from a cryomodule

Field emission (FE) from the niobium wall of a cry-omodule originates from particulates, chemical residue,and surface imperfections [18]. The rf field pulls a small

charge from an emitting site but repeats in every 1.3-GHzperiod. The resulting loss current can be sufficient toactivate or damage the module and to produce radiationnearby. Depending on their phase at emission, FE electronscan strike a nearby wall or be accelerated and transported ineither direction to other cavities or cryomodules. The LCLSNC linac is not suitable for testing the ability of the LBLMfibers to detect such losses. Instead, an experiment [19] wasperformed on a SC electron accelerator, CEBAF, at ThomasJefferson National Accelerator Facility.A 12-m fiber and two diamond detectors were tested on

the last cryomodule (1L26, a C100 high-gradient type) ofthe CEBAF north linac. The full linac had rf power, but thephotocathode laser was off, and so the only current was dueto field emission from the cavities in the cryomodules. Thefiber was draped along the side of the 10-m module. Duringthe measurements in Fig. 16, a high-radiation (B4) dia-mond and a CEBAF beam-loss monitor (referring to a PMTwith radiation-driven electron emission from the photo-cathode) were placed at the warm sections at either end ofthe cryomodule.Figure 16(a) shows the field gradients in seven of the

eight cavities of 1L26 increasing sequentially in steps of1 MeV=m (except cavity 8, which was held constant). Theother cryomodules were at full field. Two noteworthyevents occurred: Cavity 7 spiked up at t ¼ 16 min, andcavity 3 briefly dropped out at t ¼ 81 min.Figure 16(b) compares the loss detectors on the same

time axis. The upstream BLM and diamond were domi-nated by losses flowing from the previous module (1L25);their signals grew weakly with the increase in gradient andbarely registered the events in cavities 3and 7. The down-stream BLM and diamond were at the end of the linac andso did not receive losses from a neighboring cryomodule.

FIG. 15. Cividec single-crystal B1 (green line) and a poly-crystalline B2 (blue line) detectors measure a 1-W loss on anLCLS collimator as the beam rate was toggled between 0 and120 Hz. The B2 exhibits a long decay tail. The red signal is theheartbeat amplitude detected by the DSP.

FIG. 16. (a) Gradients in the eight cavities of the cryomodule. The gradients, except in cavity 8, were stepped sequentially by1 MV=m. Cavity 7 had an upward spike at t ¼ 16. Cavity 3 briefly dropped out at t ¼ 81. (b) Losses from field emission picked up byfive detectors. Only the fiber was sensitive to the sudden changes in both cavities 3 and 7.

ALAN S. FISHER et al. PHYS. REV. ACCEL. BEAMS 23, 082802 (2020)

082802-14

They showed the rf-gradient steps, although with noise(BLM) or a weak response (B4). They responded to thenearby event, in cavity 7, but did not see the drop-out incavity 3.The fiber responded only to 1L26 and was highly

sensitive to losses from its interior cavities. It showedstrong growth in field emission with each step in cavity 7,especially at the highest gradients, and jumped up with thatcavity’s upward spike. It decreased, but less sharply, for thedrop-out of cavity 3, indicating that the total signal wasdominated by cavity 7. The last upward step of cavity 7, att ¼ 91, caused a large jump in the fiber signal, and thesubsequent gradient reduction by 0.5 MV=m, at t ¼ 94,reduced the loss signal sharply. We see that the fiber issensitive to this level of field emission. The source can belocated by scanning cavity gradients.The cryomodules of the LCLS SC linac are welded

together, without intervening warm sections (except atthe bunch compressors), which at CEBAF provideperiodic access for loss detectors. A fiber parallel tothe beam just outside the cryomodules provides asensitive alternative.

E. Wire scanner

Previously, LCLS used dedicated, large-cross-sectionplastic fibers to detect losses during wire scans. Suchfibers are known to blacken with radiation and so areunsuited for use with the SC linac. Moreover, it isinefficient to introduce additional fibers, PMTs, andintegrating electronics if a good bunch profile can beobtained by sharing the diagnostic output of the newfibers. A 100-m length of the new fiber was installed1.25 m from the LCLS NC beam, beginning at a wirescanner. Figure 17 shows a test of a wire scan madeusing this new fiber, an uncooled PMT, and the oldLCLS wire-scanner electronics.

VII. SUMMARY

The LCLS-II project at SLAC adds a new superconduct-ing linac and replaces the original undulator of the x-rayFEL with two new undulators, for hard and soft x rays.The SC linac, now being installed, will have continuous rf,a bunch rate of up to 1 MHz, and up to 120 kW of beampower. This motivated a reexamination of beam-lossdetection and protection systems. The previous loss mon-itors, based on ionization chambers, have the potential tobecome nonlinear at high loss rates.Newer technologies have been selected. Key points of

concern for beam loss will be protected locally withdiamond detectors, which act as solid-state ionizationdetectors making electron-hole pairs. The entire lengthof the machine will be spanned in segments by radiation-hard optical fibers. A loss shower emits Cherenkovradiation when passing through the fiber; this light istransmitted to a PMT at one end. Both detector types willbe continuously tested by a low-level modulation at 0.8 Hz.Both detector types have been subjected to extensive

testing to qualify them as components in critical safetysystems for protection of personnel, radiation safety equip-ment, the accelerator, and the beam lines.

ACKNOWLEDGMENT

This work was supported by the U.S. Department ofEnergy, Office of Science, under Contract No. DE-AC02-76SF00515.

[1] P. Emma et al., First lasing and operation of an ångstrom-wavelength free-electron laser, Nat. Photonics 4, 641(2010).

[2] LCLS-II final design report, SLAC Report No. LCLSII-1.1-DR-0251-R0, 2015.

[3] V. Yakimenko et al., FACET-II facility for advancedaccelerator experimental tests, Phys. Rev. Accel. Beams22, 101301 (2019).

[4] J. J. Welch, Radiation detection requirements for the BCSand MPS, LCLS-II physics requirements document, SLACReport No. LCLSII-2.4-PR-0868-R1, 2018.

[5] T. N. Constant, R. W. Zdarko, R. W. Simmons, and B. M.Bennett, LIONs at the Stanford Linear Accelerator Center,SLAC Report No. SLAC-PUB-7691, 1998.

[6] R. M. Zwaska et al., Beam tests of ionization chambers forthe NuMI neutrino beam, IEEE Trans. Nucl. Sci. 50, 1129(2003).

[7] M. Santana Leitner, Fluence to dose conversion E-curvesfor silicon and polyethylene. A FLUKA user-routine toconvert fluence into energy deposition in small radiationsensitive accelerator components, SLAC radiation physicsNote No. RP-14-20, example 3, doses near a stopper.

[8] I. Dumanoglu et al., Radiation-hardness studies of highOH− content quartz fibres irradiated with 500-MeVelectrons, Nucl. Instrum. Methods Phys. Res., Sect. A490, 444 (2002).

FIG. 17. Wire scan of the LCLS vertical beam profile duringtests of the quartz fiber.

BEAM-LOSS DETECTION FOR THE HIGH-RATE … PHYS. REV. ACCEL. BEAMS 23, 082802 (2020)

082802-15

[9] K. Cankoçak, M. N. Bakırcı, S. Çerçi, E. Gülmez, J. P.Merlo, Y. Onel, F. Özok, I. Schmidt, and N. Sönmez,Radiation-hardness measurements of high OH− contentquartz fibres irradiatedwith 24-GeVprotonsup to 1.25Grad,Nucl. Instrum. Methods Phys. Res., Sect. A 585, 20 (2008).

[10] P. G. Bruecken et al., Results from post situ quartz fiberneutron irradiations, CMS Internal Note No. CMS-IN-2006/000, 2006.

[11] J. D. Jackson, Classical Electrodynamics, 3rd ed.(John Wiley and Sons, New York, 1999), Sec. 13.4.

[12] http://photocool.com/tocool.htm.[13] M. Santana-Leitner et al., Monte Carlo optimization of fast

beam loss monitors for LCLS-II, in Proceedings of theInternational Particle Accelerator Conference (IPAC’19),Melbourne, Australia (JACoW Publishing, Geneva,Switzerland, 2019), THPRB102, pp. 4066–4069, https://doi.org/10.18429/JACoW-IPAC2019-THPRB102.

[14] https://cividec.at/.[15] C. Weiss, M. Cerv, E. Griesmayer, and P. Kavrigin, Particle

interactions with diamond detectors, in Proceedingsof the International Beam Instrumentation Conference(IBIC’19), Malmö, Sweden (JACoW Publishing, Geneva,Switzerland, 2019), pp. 115–117, MOPP016, https://doi.org/10.18429/JACoW-IBIC2019-MOPP016.

[16] R. Kadyrov et al., Electronics for LCLS-II beam contain-ment system loss monitors, in Proceedings of theInternational Conference on Accelerator and LargeExperimental Physics Control Systems (ICALEPCS’19),New York, NY (JACoW Publishing, Geneva, Switzerland,2019), THCPR07, http://icalepcs2019.vrws.de/papers/thcpr07.pdf.

[17] E. Griesmayer, Cividec Instrumentation (privatecommunication).

[18] R. L. Geng, A. Freyberger, R. Legg, R. Suleiman, andA. S.Fisher, Field emission in SRF accelerators: Instrumentedmeasurements for its understanding and mitigation, inProceedings of the International Beam InstrumentationConference (IBIC’17), Grand Rapids, MI (JACoWPublishing, Geneva, Switzerland, 2017), TH1AB1,pp. 470–477, https://doi.org/10.18429/JACoW-IBIC2017-TH1AB1.

[19] A. S. Fisher et al., Beam containment and machineprotection for LCLS-2, in Proceedings of the InternationalBeam Instrumentation Conference (IBIC’17), GrandRapids, MI (JACoW Publishing, Geneva, Switzerland,2017), pp. 478–481, TH1AB2, https://doi.org/10.18429/JACoW-IBIC2017-TH1AB2.

ALAN S. FISHER et al. PHYS. REV. ACCEL. BEAMS 23, 082802 (2020)

082802-16